Title: Crop yield estimation using sentinel-2 data: a case study from Ugtaaltsaidam sub-province, Mongolia

Final project
Year of publication:
Supervisors: Michaela Hrabalikova
remote sensing, vegetation indices, correlation, crop yield, Mongolia


Globally, remote sensing technology is widely used in cropland monitoring. Even in Mongolia, there is a need to use remote sensing technology to monitor agricultural fields located in a vast area. In this study, the relationship between NDVI, NDMI, and LAI calculated using the 2017 field measurement data and Sentinel 2 satellite data of Ugtaaltsaidam sum cultivation area of Central Province was evaluated. Exploratory data analysis and land cover classification methods were used to evaluate the descriptive statistics of the collected data using QGIS Semi-Automatic Plugin. In addition, Python was used in QGIS to calculate correlations. The three calculated indices had a positive but very weak correlation with crop yield. Since there was yield data but no crop type data, unsupervised classification of land cover was done and divided into four crop types based on which were obtained by the clustering (k-mean) of Sentinel 2 image. When calculating the correlation between each of these, the three indices were positively and weakly correlated for the Crop 1 category. However, the yield of Crop 2 category was positively and weakly correlated with NDVI and LAI, but not correlated with NDMI. However, other parameters of the soil had a weak negative correlation with the indices. The results were opposite to expectations based on the literature review. The results indicate that the crop yield recording methodology has to be improved by adding information about the method of crop yield estimation, crop type, and more information about the date of harvest. The methods developed and tested in this study will be of great importance for future research on crop yield in Mongolia and will hopefully enable enhanced data collection, classification, and evaluation.

Documents and links